https://ooni.org/blog/

OONI monitors internet censorship and other forms of network interference, especially by state actors, worldwide. It’s an important contributor to digital rights and freedoms IMO, and you can run their client in the background to contribute non-personal data on pretty much any device.

How is your network handling telemetry shenanigans?

When they send over these time travelers, they’re not sending their best…

Lol why is Aqua Net hairspray specifically the thing that takes them down? Like other brands of hairspray don’t work? Spray paint or Elmer’s glue or maple syrup aren’t effective? You can’t just throw a bedsheet over them or hit them with a crowbar?

Okay but how does starting a secure shell help?

Right! For music, I think it’s even like saying… The process of making music is much more than just literally performing it… But it’d be weird for the creative process to not contain any playing-of-music that looked in some ways like performance.

I don’t think it’s just marketing bullshit to think of LLMs as AI… The research community generally does, too. Like the AI section on arxiv is usually where you find LLM papers, for example.

That’s not like a crazy hype claim like the “AGI” thing, either… It doesn’t suggest sentience or consciousness or any particular semblance of life (and I’d disagree with MW that it needs to be “human” in any way)… It’s just a technical term for systems that exhibit behaviors based on training data rather than explicit programming.

Tbh I think alot of the “thinking” still looks like visible work though. I feel like the article makes it seem a little too much like there’s nothing observable, nothing to show or demonstrate, until POOF the code comes out.

But I find that I often need to be doing visible stuff to make progress… Like devising little experiments and running them to check my assumptions about the system (or discover something new about it), and making little incremental changes, running them, using the output to guide the next thing I do… Even occasionally spending the time to write a failing test that I plan to make pass.

So I’m 100% on board with letting managers believe this “80% of the work is invisible” thing… But I think as advice for programmers, it’s really important to not get too stuck in your head and spend too much time not kinetically interacting with the system that you’re trying to change.

before getting away with swag bags full of valuables

So just look for the guy who looks like he’s just been to four different network admin conferences?

It’s actually not that high-tech… Like jamming a wifi signal is basically like just shouting over someone to prevent them from speaking (or at least from being heard). To make one from scratch, you need a little bit of technical prowess, but it’s definitely a beginner project… But to use one, you literally just turn it on, and maybe choose a frequency. They’re widely available and cheap.

There are pretty cool sophisticated digital crimes out there though, so take heart!

You’re saying you think LLMs are not AI?

An intelligent entity told us to do it. Humanity: Nah

An artificially intelligent entity told us to do it. Humanity: Well shit let’s get to work!

And this is it: Many of those “AI will be so smart that it can solve these problems for us!” arguments refer to problems where having a “smart” enough solution isn’t the problem… Getting people to care/notice/participate/get out of the way is.

My comment isn’t anti-corporate or anti-work though…? It just isn’t that strange that Google is more efficient at generating revenue (as dollars-per-kWh) than Finland is.

Opt-in is only meaningful if users can make an informed decision. I think explaining a system like PPA would be a difficult task. And most users complain a lot about these types of interruption.

In my opinion an easily discoverable opt-out option + blog posts and such were the right decision.

So you see, because the users can’t meaningfully give informed consent, their consent is therefore uh… [checks notes] not necessary.

Bullshit. Everyone knows that it’s because if you actually ask someone “do you want to be creepy tracked, less-creepy tracked, or not tracked?” they’ll pick “not” every time.

If your efficiency function is centered around revenue, then yeah, of course… No surprise that one of the world’s most successful for-profit companies generates more profit per watt-hour than a nation, which encompasses all sorts of non-revenue-generating activity like running hospitals and keeping street lights on.

Rockity rock and stone!

They should make the versions UUIDs instead of integers so that we don’t make assumptions about their ordinal relationships.

“Completely new”

Okay, then don’t train it on anything at all and let’s see how it turns out.

Warcraft: Orcs and Humans, fam.